Deep Learning and Prediction of Survival Period for Breast Cancer Patients

With the rise of deep learning, cancer-specific survival prediction is a research topic of high interest. There are many benefits to both patients and caregivers if a patient’s survival period and key factors to their survival can be acquired early in their cancer journey. In this study, we develop survival period prediction models and conduct factor analysis on data from breast cancer patients (Surveillance, Epidemiology, and End Results (SEER)). Three deep learning architectures - Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), and Recurrent Neural Networks (RNN) are selected for modeling and their performances are compared. Across both the classification and regression approaches, deep learning models significantly outperformed traditional machine learning models. For the classification approach, we obtained an 87.5% accuracy and for the regression approach, Root Mean Squared Error of 13.62% and $${R}^{2}$$ R 2 value of 0.76. Furthermore, we provide an interpretation of our deep learning models by investigating feature importance and identifying features with high importance. This approach is promising and can be used to build a baseline model utilizing early diagnosis information. Over time, the predictions can be continuously enhanced through inclusion of temporal data throughout the patient’s treatment and care.